Dashboard › institutional-transition-lab › Distillation
f1f0afdd-ec57-44f6-a9fe-bb4160ce1352["lore_tm_v1_FhDAIwZjAc-VXGm1ZtLs6nTkaTYeUd8Bskc5GnqL2z0","lore_tm_v1_-SHjnHUHelYJb2hGYOXgAF88Mk9slU2-LT-tW7E1YrU","lore_tm_v1_vBfjHdnbY17xyBineTthCgfb9N-4i19MrYmh6IfnlzY","lore_tm_v1_6QhjSHNhtcZUWz2wUpCs8bWQAVdMD73UB1eWz7OIT1M","lore_tm_v1_FJjWhNwV8T6AyGO1wGiAEOkBTJDvBHeKANdfC4YqcNw","lore_tm_v1_wjWpOvX__GIlyX9hnUUeqk8Goy8RudA9gy_NmvrLff0"]
Date: Sep 8, 2026
research/oss-governance-coding-results-v1.2.md reports frozen V1.2 results for 40 records: Luna event-class agreement 67.5%, complete power-edge agreement 55.0%, full-response agreement 0/40, provisional core agreements 5/40, routed to adjudication 35/40, valid Terra advisory responses 34/35, and canonical ledger records 0.no_event records and two product/control events; none represents an institutional power change, so every substantive institutional candidate requires source adjudication.research/oss-governance-coding-results-v1.2.md: 1. V1 recomputed a stochastic triage union and overlapped the intended 40-record envelope on only 39 records, so it is excluded from interpretation; 2. V1.1 froze exact record identity/order but its “strict edge agreement” omitted scope/evidence references and failed to route body/patch disagreement; 3. V1.2 corrected the evaluator, reused byte-identical V1.1 documents and Luna responses, and reran only Terra over the corrected route.33086454759, artifact 9653837747, digest sha256:65dba871145d26feab8eb5938a0e9ef44dc36c95215b207549d95cfe9691a6b9, used the same document-bundle SHA-256 d70f7f795847968339a645b6dbd6d4428af6bd6044de44fa2602d4c4343b0ee9.research/oss-governance-coding-results-v1.2.md: OpenTofu PR #2830 was called effective by all three coders, but they disagreed on the controlling body and emitted 3–5 edges; OpenTofu PR #727 was an effective license change for both Luna coders while Terra called it no_event; Terraform PR #33661 was recognized as an effective license change but one Luna response failed quote grounding; Valkey PRs #345 and #1390 were consistently recognized as effective steering/control-right changes but had differing exact edge sets; Valkey PR #2927 was called effective by all coders but differed on event kind, body/patch relationship, and edges, and Terra supplied one ungrounded evidence reference.33060493337, corrected continuation run 33063470968, coding artifact 9642983578, frozen-envelope commit 9ba7d77baec0b001649a4d455330a9fe298a05e9, and strict-evaluator commit 786f33a7cb044d385f49f417c059c39299f70111.research/oss-governance-coding-results-v1.2.md requires a source-level table resolving all 40 records’ event class, announcement/effective date, body-versus-patch relationship, actor, right, target, direction, scope, and evidence references; actor aliases and graph operations are canonicalized into dated power-graph diffs only after that table is frozen..github/workflows/continue-governance-coding.yml is manually dispatched, grants actions: read, contents: read, and copilot-requests: write, uses concurrency group continue-governance-coding-${{ github.ref }} with cancel-in-progress: true, and pins SOURCE_RUN_ID: "33060493337" plus source artifact oss-governance-retrieval-v2-9ba7d77baec0b001649a4d455330a9fe298a05e9.ubuntu-latest with a 30-minute timeout, actions/checkout@v7 and persist-credentials: false, Python 3.12, pinned astral-sh/setup-uv@20cfd1bf945f4377ade1205e4dbc17946fc9a30d (v10.0.1), uv sync --locked --extra analysis, Node 24, and npm ci --prefix .github/copilot-sdk --ignore-scripts --no-audit --no-fund..github/workflows/continue-governance-coding.yml re-scores the frozen files via uv run institution-lab-governance-coding compare, producing v1.2-preliminary-package.json, v1.2-terra-route.json, v1.2-preliminary-summary.json, and v1.2-preliminary-summary.md.scripts/copilot-code-governance.mjs with --input "$root/v1.2-terra-route.json", --output "$root/v1.2-terra.json", --sdk-module .github/copilot-sdk/node_modules/@github/copilot-sdk/dist/index.js, --schema schema/governance-coding-v1.schema.json, --run-id terra-advisory, --model gpt-5.6-terra, and --reasoning-effort xhigh.v1.2-adjudication-package.json, v1.2-terra-route.json, v1.2-summary.json, and v1.2-summary.md; it appends the Markdown summary to $GITHUB_STEP_SUMMARY and uploads oss-governance-coding-v1.2-${{ github.sha }} with if-no-files-found: error and retention-days: 30.scripts/copilot-code-governance.mjs permits models gpt-5.6-luna and gpt-5.6-terra, reasoning efforts low, medium, high, xhigh, and max, disables builtin:*, mcp:*, and custom:* tools, and uses TIMEOUT_MS = 180_000.scripts/copilot-code-governance.mjs are exactly entity_id, publisher, published_on, source_type, sources, and evidence_bounds; complete document keys are record_id, source_url, plus those six fields.tools: [], availableTools: [], excluded tools at session and default-agent levels, enableConfigDiscovery: false, enableSkills: false, enableSessionStore: false, infiniteSessions: { enabled: false }, memory: { enabled: false }, streaming: false, and enableSessionTelemetry: false.scripts/copilot-code-governance.mjs validates the supplied schema SHA-256 against corpus.coding_schema_sha256, requires transition_dates_excluded truthy and outcome_data_used === false, rejects empty bundles, rejects missing/extra document keys and duplicate/invalid record IDs, verifies the selected model is available, sleeps 1,000 ms between records, and requires GITHUB_TOKEN.codeOne() preserves schema_version: 1, coding_id, coding_schema_sha256, run_id, record_id, source_url, model_input_fields, model, reasoning_effort, elapsed_ms, and raw response_text.merged_at as effective_on only when the patch implements the rights change; transfers should emit removed and added edges when both are supported and use modified only if the same actor retains the right under changed terms; institutional/proposal classes require at least one grounded power change while other classes require none; every quote must match byte-for-byte; bounded omitted evidence can require abstention; source names must not be canonicalized or guessed.src/institution_lab/source_enrichment.py defines governance-relevant filename matching through RELEVANT terms governance, charter, steering, maintainer, committer, conduct, license, trademark, voting, election, committee, and foundation; patch-text matching through PATCH_TERMS additionally covers appoint, remove, vote, veto, delegate, authority, and oversight.build_audit_envelope() validates a frozen manifest with schema_version == 1, transition_dates_excluded is True, outcome_data_used is False, unique non-empty record IDs, exact record_ids_sha256, matching record_count, and selection reasons limited to narrow_query and model_review_queue; it preserves manifest order and rejects records missing from the retrieval audit.build_audit_envelope() forms the sorted union of required narrow-query hits and review-queue predictions whose decision is institutional_event or abstain, annotating each candidate with narrow_query, model_review_queue, or both._bounded_relevant_files() ranks filename matches before patch-only matches, sorts case-insensitively by filename, defaults to at most 20 files and 50,000 patch characters through enrich_envelope(), records unavailable patches separately, and marks truncation when relevant-file count or patch-character bounds are exceeded.enrich_envelope() records search_snapshot_only for non-PR sources; for PRs it captures state, draft, merged, merged_at, merge_commit_sha, base_sha, head_sha, changed_files, files_returned, files_listing_complete, pr_source_id, file_source_ids, bounded relevant_files, relevant_file_count, and patch_selection_truncated.src/institution_lab/source_enrichment.py defaults the dynamic selection rule to narrow_query_union_model_institutional_or_abstain; when an envelope manifest is supplied it records frozen_manifest:${manifest['envelope_id']}.tests/test_source_enrichment.py::test_envelope_is_fail_safe_union() expects output IDs in order ["both", "model", "narrow"], with reasons both → ["model_review_queue", "narrow_query"], model → ["model_review_queue"], and narrow → ["narrow_query"].tests/test_source_enrichment.py::test_frozen_envelope_is_independent_of_current_model_queue() verifies that a frozen manifest selecting record frozen overrides a current model queue selecting current, yielding only ["frozen"].tests/test_source_enrichment.py::test_enrichment_records_merge_state_and_bounded_relevant_patch() uses PR acme/example#7, merged at 2024-04-30T12:00:00Z, with merge/base/head SHAs merge, base, and head; it verifies 3 returned files, a complete listing, 2 relevant files, a 30-character bounded GOVERNANCE.md patch, unavailable patch status for docs/CHARTER.md, and patch_selection_truncated is True.